2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/filter.h>
25 #include <linux/skbuff.h>
26 #include <linux/vmalloc.h>
27 #include <linux/random.h>
28 #include <linux/moduleloader.h>
29 #include <asm/unaligned.h>
30 #include <linux/bpf.h>
33 #define BPF_R0 regs[BPF_REG_0]
34 #define BPF_R1 regs[BPF_REG_1]
35 #define BPF_R2 regs[BPF_REG_2]
36 #define BPF_R3 regs[BPF_REG_3]
37 #define BPF_R4 regs[BPF_REG_4]
38 #define BPF_R5 regs[BPF_REG_5]
39 #define BPF_R6 regs[BPF_REG_6]
40 #define BPF_R7 regs[BPF_REG_7]
41 #define BPF_R8 regs[BPF_REG_8]
42 #define BPF_R9 regs[BPF_REG_9]
43 #define BPF_R10 regs[BPF_REG_10]
46 #define DST regs[insn->dst_reg]
47 #define SRC regs[insn->src_reg]
48 #define FP regs[BPF_REG_FP]
49 #define ARG1 regs[BPF_REG_ARG1]
50 #define CTX regs[BPF_REG_CTX]
53 /* No hurry in this branch
55 * Exported for the bpf jit load helper.
57 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff
*skb
, int k
, unsigned int size
)
62 ptr
= skb_network_header(skb
) + k
- SKF_NET_OFF
;
63 else if (k
>= SKF_LL_OFF
)
64 ptr
= skb_mac_header(skb
) + k
- SKF_LL_OFF
;
65 if (ptr
>= skb
->head
&& ptr
+ size
<= skb_tail_pointer(skb
))
71 struct bpf_prog
*bpf_prog_alloc(unsigned int size
, gfp_t gfp_extra_flags
)
73 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
|
75 struct bpf_prog_aux
*aux
;
78 size
= round_up(size
, PAGE_SIZE
);
79 fp
= __vmalloc(size
, gfp_flags
, PAGE_KERNEL
);
83 aux
= kzalloc(sizeof(*aux
), GFP_KERNEL
| gfp_extra_flags
);
89 fp
->pages
= size
/ PAGE_SIZE
;
94 EXPORT_SYMBOL_GPL(bpf_prog_alloc
);
96 struct bpf_prog
*bpf_prog_realloc(struct bpf_prog
*fp_old
, unsigned int size
,
97 gfp_t gfp_extra_flags
)
99 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
|
103 BUG_ON(fp_old
== NULL
);
105 size
= round_up(size
, PAGE_SIZE
);
106 if (size
<= fp_old
->pages
* PAGE_SIZE
)
109 fp
= __vmalloc(size
, gfp_flags
, PAGE_KERNEL
);
111 memcpy(fp
, fp_old
, fp_old
->pages
* PAGE_SIZE
);
112 fp
->pages
= size
/ PAGE_SIZE
;
114 /* We keep fp->aux from fp_old around in the new
115 * reallocated structure.
118 __bpf_prog_free(fp_old
);
123 EXPORT_SYMBOL_GPL(bpf_prog_realloc
);
125 void __bpf_prog_free(struct bpf_prog
*fp
)
130 EXPORT_SYMBOL_GPL(__bpf_prog_free
);
132 #ifdef CONFIG_BPF_JIT
133 struct bpf_binary_header
*
134 bpf_jit_binary_alloc(unsigned int proglen
, u8
**image_ptr
,
135 unsigned int alignment
,
136 bpf_jit_fill_hole_t bpf_fill_ill_insns
)
138 struct bpf_binary_header
*hdr
;
139 unsigned int size
, hole
, start
;
141 /* Most of BPF filters are really small, but if some of them
142 * fill a page, allow at least 128 extra bytes to insert a
143 * random section of illegal instructions.
145 size
= round_up(proglen
+ sizeof(*hdr
) + 128, PAGE_SIZE
);
146 hdr
= module_alloc(size
);
150 /* Fill space with illegal/arch-dep instructions. */
151 bpf_fill_ill_insns(hdr
, size
);
153 hdr
->pages
= size
/ PAGE_SIZE
;
154 hole
= min_t(unsigned int, size
- (proglen
+ sizeof(*hdr
)),
155 PAGE_SIZE
- sizeof(*hdr
));
156 start
= (prandom_u32() % hole
) & ~(alignment
- 1);
158 /* Leave a random number of instructions before BPF code. */
159 *image_ptr
= &hdr
->image
[start
];
164 void bpf_jit_binary_free(struct bpf_binary_header
*hdr
)
166 module_free(NULL
, hdr
);
168 #endif /* CONFIG_BPF_JIT */
170 /* Base function for offset calculation. Needs to go into .text section,
171 * therefore keeping it non-static as well; will also be used by JITs
172 * anyway later on, so do not let the compiler omit it.
174 noinline u64
__bpf_call_base(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 r5
)
180 * __bpf_prog_run - run eBPF program on a given context
181 * @ctx: is the data we are operating on
182 * @insn: is the array of eBPF instructions
184 * Decode and execute eBPF instructions.
186 static unsigned int __bpf_prog_run(void *ctx
, const struct bpf_insn
*insn
)
188 u64 stack
[MAX_BPF_STACK
/ sizeof(u64
)];
189 u64 regs
[MAX_BPF_REG
], tmp
;
190 static const void *jumptable
[256] = {
191 [0 ... 255] = &&default_label
,
192 /* Now overwrite non-defaults ... */
193 /* 32 bit ALU operations */
194 [BPF_ALU
| BPF_ADD
| BPF_X
] = &&ALU_ADD_X
,
195 [BPF_ALU
| BPF_ADD
| BPF_K
] = &&ALU_ADD_K
,
196 [BPF_ALU
| BPF_SUB
| BPF_X
] = &&ALU_SUB_X
,
197 [BPF_ALU
| BPF_SUB
| BPF_K
] = &&ALU_SUB_K
,
198 [BPF_ALU
| BPF_AND
| BPF_X
] = &&ALU_AND_X
,
199 [BPF_ALU
| BPF_AND
| BPF_K
] = &&ALU_AND_K
,
200 [BPF_ALU
| BPF_OR
| BPF_X
] = &&ALU_OR_X
,
201 [BPF_ALU
| BPF_OR
| BPF_K
] = &&ALU_OR_K
,
202 [BPF_ALU
| BPF_LSH
| BPF_X
] = &&ALU_LSH_X
,
203 [BPF_ALU
| BPF_LSH
| BPF_K
] = &&ALU_LSH_K
,
204 [BPF_ALU
| BPF_RSH
| BPF_X
] = &&ALU_RSH_X
,
205 [BPF_ALU
| BPF_RSH
| BPF_K
] = &&ALU_RSH_K
,
206 [BPF_ALU
| BPF_XOR
| BPF_X
] = &&ALU_XOR_X
,
207 [BPF_ALU
| BPF_XOR
| BPF_K
] = &&ALU_XOR_K
,
208 [BPF_ALU
| BPF_MUL
| BPF_X
] = &&ALU_MUL_X
,
209 [BPF_ALU
| BPF_MUL
| BPF_K
] = &&ALU_MUL_K
,
210 [BPF_ALU
| BPF_MOV
| BPF_X
] = &&ALU_MOV_X
,
211 [BPF_ALU
| BPF_MOV
| BPF_K
] = &&ALU_MOV_K
,
212 [BPF_ALU
| BPF_DIV
| BPF_X
] = &&ALU_DIV_X
,
213 [BPF_ALU
| BPF_DIV
| BPF_K
] = &&ALU_DIV_K
,
214 [BPF_ALU
| BPF_MOD
| BPF_X
] = &&ALU_MOD_X
,
215 [BPF_ALU
| BPF_MOD
| BPF_K
] = &&ALU_MOD_K
,
216 [BPF_ALU
| BPF_NEG
] = &&ALU_NEG
,
217 [BPF_ALU
| BPF_END
| BPF_TO_BE
] = &&ALU_END_TO_BE
,
218 [BPF_ALU
| BPF_END
| BPF_TO_LE
] = &&ALU_END_TO_LE
,
219 /* 64 bit ALU operations */
220 [BPF_ALU64
| BPF_ADD
| BPF_X
] = &&ALU64_ADD_X
,
221 [BPF_ALU64
| BPF_ADD
| BPF_K
] = &&ALU64_ADD_K
,
222 [BPF_ALU64
| BPF_SUB
| BPF_X
] = &&ALU64_SUB_X
,
223 [BPF_ALU64
| BPF_SUB
| BPF_K
] = &&ALU64_SUB_K
,
224 [BPF_ALU64
| BPF_AND
| BPF_X
] = &&ALU64_AND_X
,
225 [BPF_ALU64
| BPF_AND
| BPF_K
] = &&ALU64_AND_K
,
226 [BPF_ALU64
| BPF_OR
| BPF_X
] = &&ALU64_OR_X
,
227 [BPF_ALU64
| BPF_OR
| BPF_K
] = &&ALU64_OR_K
,
228 [BPF_ALU64
| BPF_LSH
| BPF_X
] = &&ALU64_LSH_X
,
229 [BPF_ALU64
| BPF_LSH
| BPF_K
] = &&ALU64_LSH_K
,
230 [BPF_ALU64
| BPF_RSH
| BPF_X
] = &&ALU64_RSH_X
,
231 [BPF_ALU64
| BPF_RSH
| BPF_K
] = &&ALU64_RSH_K
,
232 [BPF_ALU64
| BPF_XOR
| BPF_X
] = &&ALU64_XOR_X
,
233 [BPF_ALU64
| BPF_XOR
| BPF_K
] = &&ALU64_XOR_K
,
234 [BPF_ALU64
| BPF_MUL
| BPF_X
] = &&ALU64_MUL_X
,
235 [BPF_ALU64
| BPF_MUL
| BPF_K
] = &&ALU64_MUL_K
,
236 [BPF_ALU64
| BPF_MOV
| BPF_X
] = &&ALU64_MOV_X
,
237 [BPF_ALU64
| BPF_MOV
| BPF_K
] = &&ALU64_MOV_K
,
238 [BPF_ALU64
| BPF_ARSH
| BPF_X
] = &&ALU64_ARSH_X
,
239 [BPF_ALU64
| BPF_ARSH
| BPF_K
] = &&ALU64_ARSH_K
,
240 [BPF_ALU64
| BPF_DIV
| BPF_X
] = &&ALU64_DIV_X
,
241 [BPF_ALU64
| BPF_DIV
| BPF_K
] = &&ALU64_DIV_K
,
242 [BPF_ALU64
| BPF_MOD
| BPF_X
] = &&ALU64_MOD_X
,
243 [BPF_ALU64
| BPF_MOD
| BPF_K
] = &&ALU64_MOD_K
,
244 [BPF_ALU64
| BPF_NEG
] = &&ALU64_NEG
,
245 /* Call instruction */
246 [BPF_JMP
| BPF_CALL
] = &&JMP_CALL
,
248 [BPF_JMP
| BPF_JA
] = &&JMP_JA
,
249 [BPF_JMP
| BPF_JEQ
| BPF_X
] = &&JMP_JEQ_X
,
250 [BPF_JMP
| BPF_JEQ
| BPF_K
] = &&JMP_JEQ_K
,
251 [BPF_JMP
| BPF_JNE
| BPF_X
] = &&JMP_JNE_X
,
252 [BPF_JMP
| BPF_JNE
| BPF_K
] = &&JMP_JNE_K
,
253 [BPF_JMP
| BPF_JGT
| BPF_X
] = &&JMP_JGT_X
,
254 [BPF_JMP
| BPF_JGT
| BPF_K
] = &&JMP_JGT_K
,
255 [BPF_JMP
| BPF_JGE
| BPF_X
] = &&JMP_JGE_X
,
256 [BPF_JMP
| BPF_JGE
| BPF_K
] = &&JMP_JGE_K
,
257 [BPF_JMP
| BPF_JSGT
| BPF_X
] = &&JMP_JSGT_X
,
258 [BPF_JMP
| BPF_JSGT
| BPF_K
] = &&JMP_JSGT_K
,
259 [BPF_JMP
| BPF_JSGE
| BPF_X
] = &&JMP_JSGE_X
,
260 [BPF_JMP
| BPF_JSGE
| BPF_K
] = &&JMP_JSGE_K
,
261 [BPF_JMP
| BPF_JSET
| BPF_X
] = &&JMP_JSET_X
,
262 [BPF_JMP
| BPF_JSET
| BPF_K
] = &&JMP_JSET_K
,
264 [BPF_JMP
| BPF_EXIT
] = &&JMP_EXIT
,
265 /* Store instructions */
266 [BPF_STX
| BPF_MEM
| BPF_B
] = &&STX_MEM_B
,
267 [BPF_STX
| BPF_MEM
| BPF_H
] = &&STX_MEM_H
,
268 [BPF_STX
| BPF_MEM
| BPF_W
] = &&STX_MEM_W
,
269 [BPF_STX
| BPF_MEM
| BPF_DW
] = &&STX_MEM_DW
,
270 [BPF_STX
| BPF_XADD
| BPF_W
] = &&STX_XADD_W
,
271 [BPF_STX
| BPF_XADD
| BPF_DW
] = &&STX_XADD_DW
,
272 [BPF_ST
| BPF_MEM
| BPF_B
] = &&ST_MEM_B
,
273 [BPF_ST
| BPF_MEM
| BPF_H
] = &&ST_MEM_H
,
274 [BPF_ST
| BPF_MEM
| BPF_W
] = &&ST_MEM_W
,
275 [BPF_ST
| BPF_MEM
| BPF_DW
] = &&ST_MEM_DW
,
276 /* Load instructions */
277 [BPF_LDX
| BPF_MEM
| BPF_B
] = &&LDX_MEM_B
,
278 [BPF_LDX
| BPF_MEM
| BPF_H
] = &&LDX_MEM_H
,
279 [BPF_LDX
| BPF_MEM
| BPF_W
] = &&LDX_MEM_W
,
280 [BPF_LDX
| BPF_MEM
| BPF_DW
] = &&LDX_MEM_DW
,
281 [BPF_LD
| BPF_ABS
| BPF_W
] = &&LD_ABS_W
,
282 [BPF_LD
| BPF_ABS
| BPF_H
] = &&LD_ABS_H
,
283 [BPF_LD
| BPF_ABS
| BPF_B
] = &&LD_ABS_B
,
284 [BPF_LD
| BPF_IND
| BPF_W
] = &&LD_IND_W
,
285 [BPF_LD
| BPF_IND
| BPF_H
] = &&LD_IND_H
,
286 [BPF_LD
| BPF_IND
| BPF_B
] = &&LD_IND_B
,
287 [BPF_LD
| BPF_IMM
| BPF_DW
] = &&LD_IMM_DW
,
292 #define CONT ({ insn++; goto select_insn; })
293 #define CONT_JMP ({ insn++; goto select_insn; })
295 FP
= (u64
) (unsigned long) &stack
[ARRAY_SIZE(stack
)];
296 ARG1
= (u64
) (unsigned long) ctx
;
298 /* Registers used in classic BPF programs need to be reset first. */
303 goto *jumptable
[insn
->code
];
306 #define ALU(OPCODE, OP) \
307 ALU64_##OPCODE##_X: \
311 DST = (u32) DST OP (u32) SRC; \
313 ALU64_##OPCODE##_K: \
317 DST = (u32) DST OP (u32) IMM; \
348 DST
= (u64
) (u32
) insn
[0].imm
| ((u64
) (u32
) insn
[1].imm
) << 32;
352 (*(s64
*) &DST
) >>= SRC
;
355 (*(s64
*) &DST
) >>= IMM
;
358 if (unlikely(SRC
== 0))
361 DST
= do_div(tmp
, SRC
);
364 if (unlikely(SRC
== 0))
367 DST
= do_div(tmp
, (u32
) SRC
);
371 DST
= do_div(tmp
, IMM
);
375 DST
= do_div(tmp
, (u32
) IMM
);
378 if (unlikely(SRC
== 0))
383 if (unlikely(SRC
== 0))
386 do_div(tmp
, (u32
) SRC
);
394 do_div(tmp
, (u32
) IMM
);
400 DST
= (__force u16
) cpu_to_be16(DST
);
403 DST
= (__force u32
) cpu_to_be32(DST
);
406 DST
= (__force u64
) cpu_to_be64(DST
);
413 DST
= (__force u16
) cpu_to_le16(DST
);
416 DST
= (__force u32
) cpu_to_le32(DST
);
419 DST
= (__force u64
) cpu_to_le64(DST
);
426 /* Function call scratches BPF_R1-BPF_R5 registers,
427 * preserves BPF_R6-BPF_R9, and stores return value
430 BPF_R0
= (__bpf_call_base
+ insn
->imm
)(BPF_R1
, BPF_R2
, BPF_R3
,
487 if (((s64
) DST
) > ((s64
) SRC
)) {
493 if (((s64
) DST
) > ((s64
) IMM
)) {
499 if (((s64
) DST
) >= ((s64
) SRC
)) {
505 if (((s64
) DST
) >= ((s64
) IMM
)) {
525 /* STX and ST and LDX*/
526 #define LDST(SIZEOP, SIZE) \
528 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
531 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
534 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
542 STX_XADD_W
: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
543 atomic_add((u32
) SRC
, (atomic_t
*)(unsigned long)
546 STX_XADD_DW
: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
547 atomic64_add((u64
) SRC
, (atomic64_t
*)(unsigned long)
550 LD_ABS_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
553 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
554 * only appearing in the programs where ctx ==
555 * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
556 * == BPF_R6, bpf_convert_filter() saves it in BPF_R6,
557 * internal BPF verifier will check that BPF_R6 ==
560 * BPF_ABS and BPF_IND are wrappers of function calls,
561 * so they scratch BPF_R1-BPF_R5 registers, preserve
562 * BPF_R6-BPF_R9, and store return value into BPF_R0.
565 * ctx == skb == BPF_R6 == CTX
568 * SRC == any register
569 * IMM == 32-bit immediate
572 * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
575 ptr
= bpf_load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 4, &tmp
);
576 if (likely(ptr
!= NULL
)) {
577 BPF_R0
= get_unaligned_be32(ptr
);
582 LD_ABS_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
585 ptr
= bpf_load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 2, &tmp
);
586 if (likely(ptr
!= NULL
)) {
587 BPF_R0
= get_unaligned_be16(ptr
);
592 LD_ABS_B
: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
595 ptr
= bpf_load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 1, &tmp
);
596 if (likely(ptr
!= NULL
)) {
602 LD_IND_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
605 LD_IND_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
608 LD_IND_B
: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
613 /* If we ever reach this, we have a bug somewhere. */
614 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn
->code
);
618 void __weak
bpf_int_jit_compile(struct bpf_prog
*prog
)
623 * bpf_prog_select_runtime - select execution runtime for BPF program
624 * @fp: bpf_prog populated with internal BPF program
626 * try to JIT internal BPF program, if JIT is not available select interpreter
627 * BPF program will be executed via BPF_PROG_RUN() macro
629 void bpf_prog_select_runtime(struct bpf_prog
*fp
)
631 fp
->bpf_func
= (void *) __bpf_prog_run
;
633 /* Probe if internal BPF can be JITed */
634 bpf_int_jit_compile(fp
);
635 /* Lock whole bpf_prog as read-only */
636 bpf_prog_lock_ro(fp
);
638 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime
);
640 static void bpf_prog_free_deferred(struct work_struct
*work
)
642 struct bpf_prog_aux
*aux
;
644 aux
= container_of(work
, struct bpf_prog_aux
, work
);
645 bpf_jit_free(aux
->prog
);
648 /* Free internal BPF program */
649 void bpf_prog_free(struct bpf_prog
*fp
)
651 struct bpf_prog_aux
*aux
= fp
->aux
;
653 INIT_WORK(&aux
->work
, bpf_prog_free_deferred
);
655 schedule_work(&aux
->work
);
657 EXPORT_SYMBOL_GPL(bpf_prog_free
);
659 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
660 * skb_copy_bits(), so provide a weak definition of it for NET-less config.
662 int __weak
skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
,